An updated catalog of prostate cancer predictive tools

Abstract

Accurate estimates of risk are essential for physicians if they are to recommend a specific management to patients with prostate cancer. Accurate risk estimates also are required for clinical trial design to ensure that homogeneous, high-risk patient groups are used to investigate new cancer therapeutics. Using the MEDLINE database, a literature search was performed on prostate cancer predictive tools from January 1966 to July 2007. The authors recorded input variables, the prediction form, the number of patients used to develop prediction tools, the outcome being predicted, prediction tool-specific features, predictive accuracy, and whether validation was performed. Each prediction tool was classified into patient clinical disease state and the outcome being predicted. First, the authors described the criteria for evaluation (predictive accuracy, calibration, generalizability, head-to-head comparison, and level of complexity) and the limitations of current predictive tools. The literature search generated 109 published prediction tools, including only 68 that had undergone validation. An increasing number of predictive tools addressed important endpoints, such as disease recurrence, metastasis, and survival. Despite their limitations and the limitations of data, predictive tools are essential for individualized, evidence-based medical decision making. Moreover, the authors recommend wider adoption of risk-prediction models in the design and implementation of clinical trials. Among prediction tools, nomograms provide superior, individualized, disease-related risk estimations that facilitate management-related decisions. Nevertheless, many more predictive tools, comparisons between them, and improvements to existing tools are needed. Cancer 2008. © 2008 American Cancer Society.

In the US, prostate cancer (PCa) is the most common solid malignancy and the second leading cause of cancer death among men.1 Accurate estimates of the likelihood of cancer diagnosis, stage, clinical significance, treatment success, complications, and long-term morbidity are essential for patient counseling and informed decision making. Toward this objective, properly informing the patient of these likelihoods will improve his satisfaction after treatment. Lack of patient involvement has been identified as a major risk factor for regret of treatment choice,2 particularly when complications arise.3 Therefore, accurate estimates of risk are essential for physicians if they are to recommend a specific management. Accurate risk estimates also are required for clinical trial design to ensure that homogeneous patient groups will be used to investigate new cancer therapeutics.

Traditionally, physician judgment has formed the basis for risk estimation, patient counseling, and decision making. However, clinicians' estimates may be biased because of subjective and objective confounders that exist at all stages of the prediction process.4–7 To obviate this problem and to obtain more accurate predictions, researchers have developed predictive tools (directed at predicting the probability of an outcome without considering the effect of time) and prognostic tools (directed at predicting the probability of an outcome over time) that are based on statistical models.8 Recently, predictive and prognostic nomograms have been introduced to complement the standard modeling techniques with the ability to predict the risk of the outcome of interest for the individual patient, a quality that the standard regression techniques could not offer.9 The ability of the nomograms to predict PCa diagnosis, stage, and prognosis has been confirmed.10, 11 In general, it has been demonstrated that these predictive models perform better than clinical judgment when predicting probabilities of outcome.7, 12, 13 That said, physician input obviously is essential and crucial for the measurement of variables that are used in the prediction process, for the entire decision-making process, and in the interpretation and application of model-derived outcome predictions in clinical practice.

Decision aids include the ‘Kattan-type’ nomograms,10, 11 risk groupings,14–17 artificial neural networks (ANNs),18 probability tables (such as the most widely known and applied ‘Partin staging tables’19, 20), and classification and regression tree (CART) analyses.21, 22 In 2001, we published a catalogue of PCa nomograms.8 Within the last 6 years, the field of predictive modeling has exploded, and several other valuable tools are now available.

In this review, we describe criteria for the evaluation of predictive tools and then catalogue the current nomograms available for PCa according to the natural history of treated PCa. Using MEDLINE, a literature search was performed on PCa predictive and prognostic tools from January 1966 to July 2007. For each of the identified tools, we describe the patient population and the outcomes predicted, and we record the individual characteristics of each tool, such as predictor variables, tool-specific features, predictive accuracy estimates, and whether internal and/or external validation has been performed. This review may serve as an initial step toward a comprehensive reference guide for physicians to locate published nomograms that apply to the clinical decision at hand.

Evaluating Predictive Tools

Decision aids consist of the nomograms,10, 11 risk groupings,14–17 ANNs,18 probability tables such as the ‘Partin staging tables,’19, 20 and CART analyses.21, 22 Despite the apparent differences of these prediction tools, their characteristics can be compared using a common approach. The points of comparisons are based on 4 characteristics: predictive accuracy, calibration (correlation between predicted and observed risk throughout the entire range of predictions), generalizability, and level of complexity.

Predictive accuracy

Accuracy quantifies the model's ability to discriminate between patients with and without the outcome of interest. The model's accuracy represents the most important consideration for comparison of different models. Valid determination of a model's accuracy requires application of the model under novel testing conditions different from the development cohort. Accuracy ideally should be ideally in an independent cohort. In the absence of an external cohort, models can be subjected to internal validation. Bootstrapping represents the ideal internal validation format in which the development dataset is used to simulate model testing under novel conditions.23–28 Split-sample and cross-validation (leave-1-out validation) formats represent alternatives.27

Calibration

A model's accuracy (discrimination) indicates the overall ability to predict the outcome of interest. However, overall accuracy does not indicate the ability of the model to predict the outcome of interest in specific patient groups or according to risk level. For example, a model that is 80% accurate may predict virtually perfectly well in high-risk patients but may demonstrate dismal performance in low-risk patients. The relation between predicted risk and observed rate of the outcome of interest should be provided for each new model along with its overall accuracy. Calibration plots provide this type of information and can be obtained for both internal and external data.24, 26–29 Such plots graphically illustrate the relation between predicted and observed rates of the outcome of interest. Ideally, a model with perfect ability to predict the outcome of interest should exhibit a 1:1 relation between predicted and observed rates, which results in a 45-degree slope.

Generalizability

Several considerations may undermine the generalizability of a model under specific conditions or in a specific population. For example, because of differences in the patterns of early detection and in the extent of screening, the characteristics of newly diagnosed PCa may not be the same across populations.30 These population differences may undermine the accuracy of predictive and prognostic models. Moreover, models may perform better in patients who share a specific characteristic (such as race) but may demonstrate significantly worse performance characteristics in other patients. Therefore, it is imperative for the clinician to know whether a specific model indeed is generalizable and applicable to the target population.24, 26–29

Level of complexity

The level of complexity of a predictive or prognostic model represents an important practical consideration. Excessively complex models, those that rely on multiple variables, clearly are impractical in busy clinical practice. Similarly, models that rely on variables that are not available routinely are impractical. These include models that rely on novel biomarkers, which, in turn, require noncommercially available and/or nonstandardized assays. Finally, models that require tedious preliminary calculations, for example, the calculation of prostate-specific antigen (PSA) doubling time (PSADT) without providing an ancillary PSADT calculator, tend to discourage clinicians. Therefore, parsimony represents an important ingredient. Its absence may undermine even the most accurate and best performing model.

Head-to-head comparison

When judging a new tool, its predictive accuracy, validity, and calibration should be examined relative to established models, with the intent of determining whether the new model offers advantages relative to available alternatives.9, 25, 27, 28, 31–34 Head-to-head comparisons represent the most direct and unbiased comparison of objective attributes (accuracy and calibration) of various models. Subsequently, complexity, generalizability, and other considerations can be compared. With this approach, the alternatives are compared directly without having to judge the concordance index in isolation or against a possibly arbitrary threshold.

The main steps required in a head-to-head comparison consist of the application of the original model to a common external dataset that will serve to test all models, which will be compared with one another. The original model is then applied to each individual observation to derive the probability of the outcome of interest. The predictions are then compared against observed rates of the outcome of interest, and accuracy (discrimination) is calculated using the receiver operating characteristics curve or another measure of discrimination, such as the Brier index. These steps are repeated for each of the tested models. A common mistake consists of refitting a new model that relies on the same variables as the original model and calling it the original model.

Nomograms currently represent the most accurate and discriminating tools for predicting outcomes in patients with PCa. Various studies have documented the superior performance of nomograms compared with risk grouping,9, 25, 30–32, 35, 36 look-up tables,9, 25, 32, 37, 38 tree analysis,9, 25, 39 and ANNs.9, 25, 34, 40–42 Therefore, the discussion below is weighted toward nomograms.

Limitations of Predictive and Prognostic Tools

In addition to obvious limitations related to accuracy, performance characteristics, generalizability and the level of complexity, the most common potential additional limitations of currently available predictive and prognostic tools may be classified in 1 or several of the following categories.

Study selection criteria

Specific model criteria, such as inclusion and exclusion criteria, do not allow the use of models for patients with different characteristics or who have been exposed to different treatment modalities. Therefore, it is imperative for the clinician to know whether a specific model indeed is generalizable to the target population.24, 26–29

A models' ability to predict may be affected by population characteristics that change over time. In general, more contemporary cancer patients are diagnosed with more favorable stage and grade. In consequence, tools require periodic reappraisals to assess the effect of stage and grade migration. Often, such studies may indicate that predictions devised on historic patients no longer apply to contemporary patients. However, models also may show stable accuracy and performance characteristics. External validation in contemporary cohorts is necessary to ensure temporal validity.

Adjustment for competing risks

Because of the protracted course of PCa, competing causes of mortality are extremely important in this patient population.43 Competing risks modeling can predict cancer control rates after accounting for the effects of competing risks. There is a need for more competing risks-based modeling to better situate the risk of PCa in the framework of other cause mortality.43 Such predictions are important to both clinicians and patients, especially when over treatment or suboptimal treatment considerations are addressed.

With regard to competing risks and the risk of PCa mortality, there is growing evidence indicating that the anticipated survival benefit derived from the diagnosis and treatment for PCa is nonuniform. Because the morbidity and mortality of PCa treatment are nontrivial, clinicians must be able to stratify patients PCa better according to risk, such that those who stand to gain the most from the intervention receive it. To our knowledge, to date, the only modeling tools that allow adjustment for competing risks are the Kattan-type nomogram and the Albertsen tables.43–46 Although the preliminary format of several other competing risk nomograms has been presented, to our knowledge, no such model has been published in the urologic literature to date.

Conditional probabilities

The updated versions of the pre- and postoperative Kattan nomograms for predicting biochemical recurrence after radical prostatectomy (RP) provide the opportunity to adjust for disease-free interval after surgery.47, 48 Unfortunately, few other tools demonstrate this important quality. Absence of adjustment for disease-free interval presents the clinician with an excessively somber estimate of cancer control over time. The latter, as expected, improves with increasing disease-free interval.

Suboptimal predictive accuracy

None of the prediction models developed to date are perfect. This is mainly because sufficiently informative predictors of the outcome of interest are not considered. Moreover, many models fail to include some of the key predictive risk factors that were demonstrated as virtually indispensable by others. Finally, even when all available variables are considered, models generally are not 100% accurate. Limitations in the accuracy of existing tools confirm that there is a need for novel biomarkers and imaging tools that are associated with the biologic behavior of PCa to enhance the predictive accuracy of current tools.33, 49, 50

Catalogue of Available Predictive Tools

Predicting prostate cancer on initial biopsy

Table 1 shows nomograms for predicting PCa on initial biopsy. Eastham et al developed the first nomogram for predicting PCa on initial biopsy in men who had a c-index of 0.75.51 Despite good accuracy, this nomogram suffers from limited generalizability. The nomogram was limited to men with suspicious digital rectal examination and serum PSA levels <4.0 ng/mL. In addition, the use of sextant biopsies further limits the applicability of this nomogram.

Table 1. Prostate Biopsy Nomograms for Predicting the Presence of Prostate Cancer in the Initial Biopsy Setting

Reference	Prediction Form	No. of Patients	Variables	Mean No. of Cores (Range)	Cancer Detection, %	Accuracy, %	Validation
PSA indicates prostate-specific antigen, DRE, digital rectal examination; TRUS, transrectal ultrasound the prostate; IPSS, International Prostate Symptom Score. * See Karakiewicz 2005.107 † Sampling density indicates the ratio of TRUS-derived total gland volume×the number of cores at biopsy.
Babaian 1998100	Risk group	151	Age, creatinine phosphokinase isoenzyme activity, prostatic acid phosphatase, PSA	6	24	74	Not performed
Eastham 199951	Probability nomogram development	700	Age, race, DRE, PSA of 0-4 ng/mL	6	9	75	Internal
Virtanen 1999101	Neural network	212	% Free PSA, DRE, heredity	Not available	25	81	Not performed
Finne 2000102	Neural network	656	% Free PSA, PSA, DRE, TRUS	Not available	23	Not available	Not performed
Horninger 2001103	Neural network	3474	Age, PSA, % free PSA, DRE, TRUS, PSA density, PSA density of transition zone, transition zone volume	Not available	Not available	Not available	Not performed
Kalra 2003104	Neural network	348	Age, ethnicity, heredity, IPSS, DRE, PSA, complexed PSA	6	Not available	83	Not performed
Garzotto 2003105	Probability nomogram development	1239	Age, race, family history, referral indications, prior vasectomy, DRE, PSA ≤10 ng/mL, PSA density, TRUS findings	6.7 (6-13)	24	73	Not performed
Finne 2004106	Neural network	1775	DRE, % free PSA, TRUS, PSA	Not available	22	76	Not performed
Karakiewicz 2005107	Probability nomogram development	6469	Age, DRE, PSA, % free PSA	6	35-42	77	Internal andexternal
Porter 2005108	Neural network	3814	Age, PSA, gland volume, PSA density, DRE, TRUS	6	27-42	72-75	Internal and external
Suzuki 2006109	Probability nomogram development	834	Age, PSA, % free PSA, prostate volume, DRE	≥6	29	82	Internal
Chun 200752	Probability nomogram validation and development*	2900	Age, DRE, PSA, % free PSA, sampling density†	11 (10-20)	41	77	Internal andexternal

Chun et al developed and validated nomograms in a population that was exposed to extended biopsy sampling.52 External validation in 3 cohorts totaling 2900 men demonstrated 73% to 76% accuracy.

Predicting prostate cancer on repeat biopsy

Table 2 shows nomograms for predicting PCa on repeat biopsy. Lopez-Corona et al developed a nomogram that predicts the probability of a positive repeat biopsy.53 The nomogram was developed and internally validated (70% accuracy) in 343 men and externally validated (71% accuracy) in 230 patients.54

Table 2. Prostate Biopsy Nomograms for Predicting the Presence of Prostate Cancer in Other Than an Initial Biopsy Setting

Reference	Prediction Form	Design	No. of Patients	Variables	Median No. of Previous Biopsy Sessions	Mean No. of Cores (Range)	Cancer Detection, %	Accuracy, %	Validation
PSA indicates prostate-specific antigen; DRE, digital rectal examination; HGPIN. high-grade intraepithelial neoplasia; ASAP, atypical small acinar proliferation of the prostate; TRUS, transrectal ultrasound of the prostate; BPH, benign prostatic hyperplasia. * Sampling density indicates the ratio of TRUS-derived total gland volume × the number of cores at biopsy.
O'Dowd 2000110	Probability nomogram Development	Repeat biopsy	813	Age, initial biopsy diagnosis, PSA, % free PSA	Not available	Not available	29	70	Not performed
Lopez-Corona 200353	Probability nomogram Development	Repeat biopsy	343	Age, DRE, no. previous negative biopsies, HGPIN history, ASAP history, PSA, PSA slope, family history, months from initial negative biopsy	2.9 (2-12)	9.2 (6-22)	20	70	Internal
Remzi 2003111	Neural network	Repeat biopsy	820	PSA, % free PSA, TRUS, PSA density, PSA density of the transition zone, transition zone volume	Not available	8	10	83	Not performed
Yanke 200554	Probability nomogram validation54	Repeat biopsy	230 (356 Biopsies)	Age, DRE, no. previous negative biopsies, HGPIN history, ASAP history, PSA, PSA slope, family history, months from initial negative biopsy, months from previous negative biopsy	2.6 (2-7)	17.9 (12-54)	34	71	Internal
Chun 200755	Probability nomogramDevelopment	Repeat biopsy	2393	Age, DRE, PSA, % free PSA, no. previous negative biopsies, sampling density*	1.5 (1-7)	11 (10-24)	30	76	Internal and external
Walz 2006112	Probability nomogram Development	Repeat saturation biopsy	161	Age, PSA, % free PSA, prostate and BPH volume, PSA doubling time, PSA density of the transition zone, no. of previous biopsy sessions, no. of cores at saturation biopsy	2.5 (2-5)	24.5 (20-32)	41	75	Internal
Snow 199418	Neural network	Initial and repeat biopsy	1787	Age, change on PSA, DRE, PSA, TRUS	Not available	6	34	87	Not performed
Carlson 1998113	Probability table	Initial and repeat biopsy	3773	Age, PSA, % free PSA	Not available	6	33	Not available	Internal
Djavan 2002114	Neural network	Initial and repeat biopsy	272	PSA density of the transition zone, % free PSA, PSA density, TRUS (PSA of 2.5-4.0 ng/mL)	Not available	8	24	88	Not performed
			974	PSA density of the transition zone, % free PSA, PSA velocity, transition zone volume, PSA, PSA density (PSA of 4.0-10.0 ng/mL)	Not available	8	35	92	Not performed
Stephan 2002115	Neural network	Initial and repeat biopsy	1188	Age, DRE, PSA, % free PSA, TRUS	Not available	Not available	61	86	Not performed
Porter 2002116	Neural network	Initial and repeat	319	Age, PSA, gland volume, TRUS, DRE, previous negative biopsy, African-American race	Not available	9.7 (6-10)	39	76	Not performed
Matsui 2004117	Neural network	Initial and repeat biopsy	228	PSA density, DRE, age, TRUS	Not available	10-12	26	73	Not performed
Benecchi 2006118	Neural network	Initial and repeat biopsy	1030	Age, PSA, % free PSA	Not available	6-12	19	80	Not performed
Yanke 2006119	Probability nomogramDevelopment	Initial and repeat biopsy	8851	Age, race, PSA, DRE, no. of cores	Not available	6-13	27-38	75	Internal

Chun et al developed the most contemporary repeat biopsy nomogram (n = 721) based on ≥10 biopsy cores.55 In an external validation cohort (n = 361), the nomogram yielded 74% accuracy.

Staging Nomograms

Several multivariate statistical models have been proposed to estimate pathologic stage at RP with the intent of facilitating treatment planning. Of these, the ‘Partin tables’ represent the most widely used tool. These look-up tables predict pathologic stage at RP (Table 3).19 After their introduction in 1993, the Partin tables were validated in 1997 and updated in 2001 and 2007.20, 56–58

Table 3. Prediction of Pathologic Stage in Men Who Underwent Radical Prostatectomy for Clinically Localized Prostate Cancer

Reference(s)	Prediction Form	Outcome	No. of Patients	Variables	Accuracy. %	Validation
PSA indicates prostate-specific antigen; TRUS, transrectal ultrasound of the prostate; RP, radical prostatectomy; LN, lymph node.
Narayan 1995120	Probability graph	Pathologic stage	813	Biopsy based stage, biopsy Gleason sum, PSA	Not available	Not performed
Partin 199719 and 1993121; Makarov 200720	Probability table	Pathologic stage	703 and 4133	Biopsy Gleason sum, clinical stage, PSA	Internal. 72; external,. 84	External (Kattan 1997122) and updated (Markov 2007,20 Partin 2001,58 Blute 2000123)
Epstein 1994124	Risk group	Clinically indolent cancer defined as pathologically organ confined, tumor volume ≤0.2 cc, Gleason sum <7	157	Biopsy Gleason sum, mm core with cancer, PSA density, no adverse pathologic findings on needle biopsy	Not available	External (Carter & Epstein 1994125)
Goto 1996126	Risk group	Clinically indolent cancer defined as pathologically organ confined, tumor volume ≤0.5 cc, Gleason sum <7	569	PSA density, maximal mm cancer in any core	Not available	Not performed
Kattan 200365	Probability nomogram development	Clinically indolent cancer defined as pathologically organ confined, tumor volume ≤0.5 cc, no Gleason grade 4 or 5	409	PSA, primary and secondary biopsy Gleason sum	64	Internal and external (Steyerberg 200729)
				PSA, primary and secondary biopsy Gleason sum, % positive cores, TRUS volume	74	Internal and external (Steyerberg 200729)
				PSA, clinical stage, primary and secondary biopsy Gleason sum, TRUS volume, mm core with cancer, mm core without cancer	79	Internal and external (Steyerberg 200729)
Chun 2006127	Probability nomogram development	Gleason upgrading between biopsy and RP	2982	PSA, clinical stage, primary and secondary biopsy Gleason sum	80	Internal
Chun 2006128	Probability nomogram development	Significant Gleason upgrading between biopsy and RP	4789	PSA, clinical stage, biopsy Gleason sum	76	Internal
Steuber 2006129	Probability nomogram development	Tumor location: transition vs peripheral zone	945	PSA, biopsy Gleason sum, positive biopsy cores at midprostate only, no. of positive biopsy cores at base, cumulative % biopsy tumor volume	77%	Internal
Peller 1995130	Probability table	Tumor volume	102	Biopsy Gleason sum, no. of positive sextant cores, PSA	Not available	Not performed
Ackerman 1993131	Probability formula	Surgical margin positivity	107	No. of positive sextant cores, PSA density	70	Not performed
Rabbani 1998132	Probability graph	Surgical margin positivity	242	Androgen deprivation, no. of ipsilateral positive cores, PSA	Not available	Not performed
Bostwick 1996133	Probability graph	Capsular penetration	314	Biopsy Gleason sum, % cancer in biopsy cores, PSA	78	Not performed
Gamito 2000134	Neural network	Capsular penetration	4133	Age, race, PSA, PSA velocity, Gleason sum, clinical stage	30-76	External
Gilliland 1999135	Probability graph	Extracapsular extension	3826	Age, biopsy Gleason sum, PSA	63	Not performed
Ohori 200459	Probability nomogram development	Side-specific extracapsular extension	763	PSA, clinical stage, side-specific biopsy Gleason sum, side-specific % positive cores, side-specific % of cancer in cores	81	External (Steuber 200639)
Steuber 200639	Probability nomogram development	Side-specific extracapsular extension	1118	PSA, clinical stage, biopsy Gleason sum, % positive cores, % of cancer in positive cores	84	Internal
Veltri 2001,136 Haese 2003137	Ordinal logistic regression and neural network	Organ confined disease	1287	Age, PSA, no. of positive cores, highest Gleason score, average % tumor involvement per core, presence of Gleason pattern 4/5, midcore with >5% tumor base and/or midcore with >5% tumor	93-98.6	External
Badalament 1996138	Probability formula	Organ confined disease	192	Biopsy Gleason sum, involvement of >5% of base with or without apex biopsy, nuclear grade, PSA, total % tumor involvement	86	Not performed
Bostwick 1996133	Probability graph	Seminal vesicle invasion	314	Biopsy Gleason sum, % cancer in cores, PSA	76	Not performed
Pisansky 1996139	Probability graph	Seminal vesicle invasion	2953	Biopsy Gleason primary grade, clinical stage, PSA	80	Internal
Koh 200360	Probability nomogram Development	Seminal vesicle invasion	763	PSA, clinical stage, primary and secondary Gleason sum, and % of cancer at the base	88	Internal
Baccala 2007140	Probability nomogram Development	Seminal vesicle invasion	6740	Age, PSA, Biopsy Gleason sum, clinical stage	80	Internal
Gallina 200738	Probability nomogram development	Seminal vesicle invasion	896	PSA, clinical stage, biopsy Gleason sum, % positive biopsy cores	79	Internal and external
Ackerman 1993131	Probability formula	LN invasion assessed with limited pelvic lymphadenectomy	107	No. of positive sextant cores, PSA	94	Not performed
Roach 1994141	Probability graph	LN invasion assessed with limited pelvic lymphadenectomy	212	Biopsy Gleason sum, PSA	Not available	Not performed
Bluestein 1994142	Probability graph	LN invasion assessed with limited pelvic lymphadenectomy	816	Biopsy Gleason sum, clinical stage, PSA	82	Internal
Batuello 2001143	Neural network	LN invasion assessed with limited pelvic lymphadenectomy	6454	Biopsy Gleason sum, clinical stage, PSA	77-81	Internal and external
Cagiannos 200361	Probability nomogram Development	LN invasion assessed with limited pelvic lymphadenectomy	5510	PSA, clinical stage, biopsy Gleason sum/PSA, clinical stage, biopsy Gleason sum, institution	76/78	Internal
Briganti 2006,37 2006,62 and 2007144	Probability nomogram development	LN invasion assessed with extended pelvic lymphadenectomy (≥10 LNs)	78137	PSA, clinical stage, biopsy Gleason sum, no. of LNs	79	Internal
			60262	PSA, clinical stage, biopsy Gleason sum	76	Internal
			278144	PSA, clinical stage, biopsy Gleason sum, % positive biopsy cores	83	Internal

Although the Partin tables represent a milestone in pretreatment PCa staging, they have limitations. For example, the probability of extracapsular extension (ECE) cannot be predicted in a side-specific fashion. To circumvent this limitation, Graefen et al attempted to enhanced the specificity of ECE predictions by devising a regression tree analysis capable of predicting ECE in a side-specific manner.22 Their model allows the identification of candidates for nonnerve-sparing versus unilateral versus bilateral nerve-sparing prostatectomy. External validation of this model yielded 70% accuracy.57

Ohori et al developed a nomogram (n = 763 patients) to predict side-specific ECE (range of c-index, 79%-81%).59 Validation of another side-specific nomogram in 1118 European patients yielded 84% accuracy.39 Compared with the Partin tables, the nomogram approach is more accurate and provides side-specific predictions. Moreover, the nomograms predict ECE independent of seminal vesicle invasion (SVI) and lymph node invasion (LNI).19, 20, 58

Koh et al60 and Gallina et al38 devised nomograms to predict the probability of SVI in 763 patients (range of c-index, 78%-88%). Cagiannos et al (n = 5510 patients) developed an LNI nomogram that yielded 76% accuracy compared with 74% for the Partin tables that were tested in the same cohort.61 Briganti et al developed an LNI nomogram in 602 patients who underwent extended pelvic lymphadenectomy (76% accuracy).62 In addition, the same investigators developed another nomogram (n = 565 patients) that allows the identification of patients who are at a negligible risk of LNI outside of the obturator fossa and in whom an extended pelvic lymphadenectomy can be omitted (80% accuracy).63 The combination of these nomograms allows accurate identification of the need and extent of pelvic lymphadenectomy.

Although predicting adverse pathologic features is important for the management of patients with PCa, a proportion of patients harbors clinically insignificant or indolent PCa that cannot be predicted with these tools.64 To address this void, Kattan et al developed 3 nomograms (range of c-index, 64%-79%) that predict the probability of indolent PCa65 based on the definition of Epstein et al (ie, organ-confined cancer, ≤0.5 mL in volume, and no poorly differentiated elements).66 Steyerberg et al externally validated those nomograms in a screening cohort (range of c-index, 61%-76%).29 These tools allow the prediction of clinically insignificant PCa with reasonable accuracy, which may help in the decision-making process between definitive therapy versus active surveillance.

Predicting Biochemical Recurrence After Radical Prostatectomy

Before radical prostatectomy

Kattan et al developed a pretreatment nomogram (n = 983 patients) that predicts 5-year biochemical recurrence for patients who undergo RP (Fig. 1) (Table 4).10 External validation yielded accuracies of ≈75% (range, 65%-83%).67–69

Figure 1. Preoperative nomogram for predicting the 5-year probability of freedom from biochemical disease recurrence after radical prostatectomy. PSA indicates prostate-specific antigen. Reproduced with permission from Kattan MW, Eastham JA, Stapleton AM, Wheeler TM, Scardino PT. A preoperative nomogram for disease recurrence following radical prostatectomy for prostate cancer. J Natl Cancer Inst. 1998;90:766-771. Reprinted with permission of Oxford University Press.

Table 4. Preoperative and Postoperative Prediction of Biochemical Recurrence in Men Who Underwent Radical Prostatectomy

Reference(s)	Prediction Form	Operative Status	Biochemical Recurrence. y	No. of Patients	Variables	Accuracy, %	Validation
PSA indicates prostate-specific antigen; LN, lymph node.
Snow 199418	Neural network	Preoperative	Not available	240	Age, PSA, clinical stage, biopsy Gleason grade, potency	90	Not performed
Kattan 199810	Probability nomogram development	Preoperative	5	983	Biopsy primary and secondary Gleason grade, clinical stage, PSA	Internal, 74; external, 65-83	Internal and external (Graefen 200267, 68 and 2003146; Greene 2004145)
D'Amico 199916	Probability table	Preoperative	2	892	Biopsy Gleason sum, clinical stage, PSA	Not available	Internal
Graefen 1999147	Probability graph	Preoperative	3.5	315	Biopsy Gleason sum, no. of positive cores, PSA	Not available	Not performed
D'Amico 200017	Probability graph	Preoperative	2	977	Biopsy Gleason sum, endorectal coil magnetic resonance imaging T-stage, PSA, % positive biopsy cores	Not available	Internal
Tewari 2001148	Neural network	Preoperative	3.5	1400	Age, race, PSA, clinical staging, biopsy Gleason sum	83	Not performed
D'Amico 1998,14 200215	Probability graph	Preoperative	4	823	Biopsy Gleason sum, clinical stage, PSA, % positive biopsy cores	80	Internal and external (Graefen 2002,67 Mitchell 2005149)
Cooperberg 2005150	Probability graph	Preoperative	3 and 5	1439	Age, PSA, biopsy Gleason sum, clinical stage, % positive biopsy	Internal, 66; external. 68-81	Internal and external (Cooperberg 2006,151 May 2007152)
Stephenson 200648	Probability nomogram development	Preoperative	10	1978 and 1545	PSA, clinical stage, biopsy Gleason sum, year of surgery, no. of positive and negative cores	76-79%	Internal and external
Bauer 1998153	Probability formula	Postoperative	5	378	Race, PSA, Gleason sum, organ confined status	Not available	External (Moul 2001154)
D'Amico 199814	Probability graph	Postoperative	2	862	Pathologic stage, PSA, Gleason sum, surgical margin status	Not available	Not performed
Graefen 1999147	Probability graph	Postoperative	3.5	318	Pathologic stage, volume Gleason grade 4/5	Not available	Not performed
Potter 1999155	Neural network	Postoperative	5	214	Gleason sum, extraprostatic extension, surgical margin status, age, DNA ploidy, and quantitative nuclear grade	94	Internal
Kattan 199911	Probability nomogram development	Postoperative	5	996	PSA, Gleason sum, extracapsular extension, seminal vesicle invasion, LN invasion, surgical margin status	Internal, 89; external, 77-83	Internal and external (Bianco 2003,69 Graefen 2002,73 Ramsden and Chodak 2004156)
Stamey 2000157	Probability formula	Postoperative	Unknown	326	PSA, % Gleason grade 4/5, volume largest cancer, vascular invasion, prostate weight, % intraductal cancer, LN invasion	Not available	Not performed
McAleer 2005158	Probability graph	Postoperative	7	2417	Gleason grade, stage, margin status, dichotomized PSA (cut point 10 ng/mL).	Not available	Internal
Stephenson 200547	Probability nomogram development	Postoperative	10	1881, 1782, and 1357	PSA, Gleason sum, extracapsular extension, seminal vesicle invasion, LN invasion, surgical margin status	78-86	Internal and external

However, the 5-year endpoint is insufficient to predict the likelihood of cure after RP, because many patients are at risk of disease recurrence beyond 5 years.70–72 Therefore, Stephenson et al recently updated the preoperative nomogram by predicting the 10-year probability of biochemical recurrence after RP (Fig. 2A) (77% accuracy).48 Their model exhibited good calibration across the spectrum of predictions in internal validation but exhibited some optimism in external validation (Fig. 2B). An added feature of the nomogram is the ability to estimate the probability of recurrence at any point in time from 1 year to10 years after RP after accounting for disease-free interval.

Figure 2. (A) Preoperative nomogram estimating the 1-year to 10-year biochemical recurrence-free probability after radical prostatectomy alone. (B) Calibration plot of the nomogram in external validation. The 45° line represents an ideal model in which estimates of disease recurrence are calibrated perfectly with outcome. Vertical bars are 95% confidence intervals for quintiles in the validation set. PSA indicates prostate-specific antigen. Reproduced with permission from Stephenson AJ, Scarcino PT, Eastham JA, et al. Preoperative nomogram predicting the 10-year probability of prostate cancer recurrence after radical prostatectomy. J Natl Cancer Inst. 2006;98:715-717. Reprinted with permission of Oxford University Press.

After radical prostatectomy

Kattan et al also developed a postoperative nomogram for predicting 5-year biochemical recurrence using data from 996 men who underwent RP for clinically localized PCa by a single surgeon (73% accuracy).11 External validation in an international cohort and in African-American men yielded accuracies of 80% (range, 77%-82%)73 and 83%,69 respectively. Stephenson et al updated the postoperative nomogram by including contemporary patients and extending predictions up to 10 years after radical surgery while accounting for disease-free interval (Fig. 3).47 External validation yielded an accuracy from 78% to 81%.47 Suardi et al developed the furthest reaching nomogram, which provides the probability of biochemical recurrence up to 20 years after RP. Their model's accuracy (77%-83%) was confirmed in 2 external validation cohorts, and the model provides conditional probabilities of biochemical recurrence after adjustment for disease-free interval.74

Figure 3. Postoperative nomogram predicting 10-year biochemical recurrence-free probability after radical prostatectomy (RP). Neg indicates negative; Pos, positive; Inv, invasion; PSA, prostate-specific antigen. Reproduced with permission from Stephenson AJ, Scardino PT, Eastham JA, et al. Postoperative nomogram predicting the 10-year probability of prostate cancer recurrence after radical prostatectomy. J Clin Oncol. 2005;23:7005-7012. Reprinted with permission. ©2008 American Society of Clinical Oncology. All rights reserved.

Predicting Biochemical Recurrence After External-beam Radiotherapy or Brachytherapy

Virtually all patients are interested in the projected efficacy of different treatment modalities. Therefore, pretreatment estimates of outcomes if patients receive either 3-dimensional, conformal, external-beam radiotherapy or brachytherapy are important in pretreatment counseling (Table 5). The original American Society for Therapeutic Radiology and Oncology definition of 3 consecutive increases in PSA for biochemical recurrence was used as the endpoint in all studies listed in Table 5.75 Kattan et al developed a pretreatment nomogram to predict the 5-year biochemical recurrence-free probability after 3-dimensional, conformal, external-beam radiotherapy (n = 1042 patients) (Fig. 4).76 External validation within a cohort of 912 men yielded an accuracy of 76%.77, 78

Figure 4. Pretreatment nomogram for predicting 5-year biochemical disease recurrence-free probability (Prob.) after 3-dimensional conformal radiotherapy. PSA indicates prostate-specific antigen; Bx, biopsy; Gy, grays. Reproduced with permission from Kattan MW, Zelefsky MJ, Kupelian PA, Scardino PT, Fuks Z, Leibel SA. Pretreatment nomogram for predicting the outcome of 3-dimensional conformal radiotherapy in prostate cancer. J Clin Oncol. 2000;18:3352-3359. Reprinted with permission. ©2008 American Society of Clinical Oncology. All rights reserved.

Table 5. Pretreatment Prediction of Biochemical Recurrence in Men Treated With Radiotherapy

Reference(s)	Prediction Form	Treatment	Outcome (Years)	No. of Patients	Variables	Accuracy, %	Validation
RT indicates radiotherapy; PSA, prostate-specific antigen; Gy, grays.
Duchesne 1996159	Risk group	External-beam RT	Biochemical recurrence (5)	85	PSA, biopsy Gleason sum	Not available	Not performed
Pisansky 1997160	Risk group	External-beam RT	Biochemical recurrence (5)	500	Biopsy Gleason sum, clinical stage, PSA	Not available	Internal
Zagars 1997161	Probability graph	External-beam RT	Biochemical recurrence (6)	938	PSA, biopsy Gleason sum, clinical stage	Not available	Not performed
D'Amico 199916	Probability table	External-beam RT	Biochemical recurrence (2)	762	Biopsy Gleason sum, clinical stage, PSA	Not available	Not performed
Shipley 1999162	Probability table	External-beam RT	Biochemical recurrence (5)	1607	Biopsy Gleason sum, clinical stage, PSA	Not available	Not performed
Kattan 200076	Probability nomogramdevelopment	External-beam RT	Biochemical recurrence (5)	1042 and 1030	PSA, biopsy Gleason sum, clinical stage, neoadjuvant androgen-deprivation therapy, radiation dose delivered	73	Internal
D'Amico 2002,77199878	Probability graph	External-beam RT	Biochemical recurrence (5)	766	Biopsy Gleason sum, clinical stage, PSA, treatment modality	Not available	Internal
D'Amico 199878	Probability graph	Brachytherapy	Biochemical recurrence (5)	218	Biopsy Gleason sum, clinical stage, PSA, neoadjuvant therapy	Not available	Not performed
Ragde 1998163	Risk group	Brachytherapy	Biochemical recurrence (10)	98	Age, biopsy Gleason sum, clinical stage, PSA, 45-Gy external-beam RT	76	Internal
Kattan 200179	Probability nomogram development	Brachytherapy	Biochemical recurrence (5)	920, 1827, and 765	Biopsy Gleason sum, clinical stage, PSA, coadministration of external-beam RT	61	Internal and external

The same authors developed a model that predicts 5-year biochemical recurrence-free survival after ¹²⁵I-seeds brachytherapy without adjuvant hormone therapy (n = 920 patients) (Fig. 5).79 Two separate external validations resulted in accuracies of 61% (n = 1827 patients) and 64% (n = 765 patients).

Figure 5. Pretreatment nomogram for predicting 5-year biochemical disease recurrence-free probability (Rec Free Prob.) after permanent prostate brachytherapy without neoadjuvant androgen ablative therapy. PSA indicates prostate-specific antigen; Gl. sum, Gleason sum; XRT, external-beam radiotherapy. Reproduced with permission from Kattan MW, Potters L, Blasko JC, et al. Pretreatment nomogram for predicting freedom from recurrence after permanent prostate brachytherapy in prostate cancer. Urology. 2001;58:393-399. ©2008, with permission from Elsevier and the Société Internationale d' Urologie.

Predicting Metastatic Disease Progression

To address metastatic progression after definitive therapy, Kattan et al developed a nomogram that quantifies the probability of metastatic progression within 5 years after external-beam radiotherapy (n = 1677 patients).80 Its external validation demonstrated an accuracy of 81% (n = 1626 patients). The tool can predict the likelihood of metastatic progression immediately after definitive therapy (Table 6).

Table 6. Prediction of Metastasis and Survival

Reference(s)	Prediction Form	Patient Population	Outcome (Years)	No. of Patients	Variables	Accuracy, %	Validation
RP indicates radical prostatectomy; PSA, prostate-specific antigen; RT, radiotherapy; LN, lymph node; LDH, lactate dehydrogenase; ALP, alkaline phosphatase; ECOG, Eastern Cooperative Oncology Group.
Partin 1994164	Probability graph	RP	Local vs distant recurrence	1058	PSA velocity, Gleason sum, pathologic stage	Not available	Not performed
Pound 199970	Probability table	Biochemical recurrence after RP	Metastasis (7 y after biochemical recurrence)	315	PSA doubling time, Gleason sum, time to biochemical recurrence	56	Not performed
D'Amico 2003165	Probability graph	RP	Prostate cancer-specific mortality (8)	4946	Biopsy Gleason sum, clinical stage, PSA	Not available	Internal
Dotan 200581	Probability nomogram development	Biochemical recurrence after RP	Positive bone scan	239	Pretreatment PSA, surgical margin status, seminal vesicle invasion, Gleason sum, trigger PSA, extracapsular extension, PSA slope, PSA velocity	93	Internal
Freedland 200571	Probability table	Biochemical recurrence after RP	Cancer-specific survival (10 y after biochemical recurrence)	379	PSA doubling time, Gleason sum, time from surgery to biochemical recurrence	59	Not performed
D'Amico 2002166 and 2003167	Probability graph	External-beam RT	Prostate cancer-specific mortality (10)	381	Biopsy Gleason sum, clinical stage, PSA, % positive biopsy	Not available	Internal
				94	Time to PSA failure, posttreatment PSA doubling time, timing of salvage hormone therapy	Not available	Internal
D'Amico 2003167	Probability graph	External-beam RT	Prostate cancer-specific mortality (8)	2370	Biopsy Gleason sum, clinical stage, PSA	Not available	Internal
Kattan 200380	Probability nomogram development	External-beam RT	Metastasis (5)	1677 and 1626	PSA, clinical stage, biopsy Gleason sum	81	Internal and external
Slovin 200582	Probability nomogram development	External-beam RT	Metastasis (1-2)	148	Baseline PSA, PSA doubling time, pathologic T classification, Gleason sum	69	Not performed
Zhou 2005168	Probability graph	External-beam RT	Prostate cancer-specific mortality (5)	661	PSA doubling time, biopsy Gleason sum	Not available	Internal
Stephenson 2007169	Probability nomogram development	Salvage RT for biochemical recurrence after RP	Biochemical recurrence after radiotherapy (7 y after biochemical recurrence)	1540	Prostatectomy PSA, Gleason sum, seminal vesicle invasion, extracapsular extension, surgical margin status, LN metastasis, persistently elevated PSA after RP, pre-RT PSA, PSA doubling time, neoadjuvant androgen-deprivation therapy, radiation dose	69	Internal
Zhou 2005168	Probability graph	Biochemical recurrence after RP	Prostate cancer-specific mortality (5)	498	PSA doubling time	Not available	Internal
Slovin 200582	Probability nomogram development	Biochemical recurrence after RP or RT	Metastasis (1-2)	148	Baseline PSA, PSA doubling time, pathologic T classification, Gleason sum	69	Not performed
Svatek 2006170	Probability nomogram development	Androgen-independent prostate cancer	Prostate cancer-specific mortality (1-5)	129	PSA at initiation of androgen-deprivation therapy, PSA doubling time, nadir PSA on androgen-deprivation therapy, time from androgen-deprivation therapy to androgen-independent prostate cancer	81	Internal
Smaletz 200284	Probability nomogram development	Men with progressive metastatic prostate cancer after castration	Overall survival (1-2)	409 and 433	Age, Karnofsky performance index, hemoglobin, PSA, LDH, ALP, albumin	71	Internal and external
Porter 200783	Probability nomogram development	Men on androgen-deprivation therapy after RP	Prostate cancer-specific mortality (2-5)	66	Pathologic T classification, Gleason sum, surgical margin status, age at androgen-deprivation therapy, recurrence type	66	Internal
Halabi 200385	Probability nomogram development	Metastatic hormone-refractory prostate cancer	Overall survival (1-2)	1101	LDH, PSA, ALP, Gleason sum, ECOG performance status, hemoglobin, presence of visceral disease	68	Internal and external

Dotan et al developed a nomogram for predicting the probability of metastatic progression, which they defined as a positive bone scan, in 239 men with a rising PSA after RP.81 The nomogram relies on detailed serum PSA characteristics, including kinetics, resulting in an accuracy of 93%. This nomogram requires multiple postrecurrence PSA values to allow the estimation and inclusion of PSA kinetics.

Slovin et al devised a similar nomogram for predicting the time to radiographically detectable metastases in patients with biochemical recurrence (n = 148) after either RP or external-beam radiotherapy. Similar to the nomogram described by Dotan et al, the nomogram described by Slovin et al requires the consideration of PSA kinetics in the form of PSADT and is limited to patients with PSADT values <.82 The predictive accuracy of this model was 69%, but it was validated neither internally nor externally.

Predicting Survival

Four nomograms have been devised for predicting survival in PCa patients (Table 6). Of these, 1 predicts cause-specific survival in PCa patients exposed to hormone therapy, regardless of the time of hormone therapy initiation. The remaining 3 nomograms predict the probability of all-cause survival in patients with androgen-insensitive PCa (AIPC).

Porter et al developed a nomogram for predicting cause-specific survival in patients who were exposed to hormone therapy after RP failure (n = 114 patients).83 The internally validated accuracy of their nomogram was only 66%. The nomograms of Smaletz et al84 and of Halabi et al85 were developed and externally validated in heavily pretreated patients with AIPC who had been exposed to 1 or several experimental agents. In external validation, the accuracies of the nomograms of Smaletz et al and Halabi et al were 67% and 68%, respectively.

Svatek et al devised a contemporary nomogram using a population with a median survival of 52 months who had not received experimental therapies. This nomogram relies predominantly on PSADT and PSA at hormone therapy initiation. Internal validation of this nomogram yielded an accuracy of 81%. The contemporaneous nature and homogeneity of the patient population make this model very attractive when survival needs to be assessed in patients with AIPC.

Predicting Life Expectancy

Life expectancy plays a crucial role in treatment considerations, especially in patients who are candidates for definitive therapy (Table 7). The 10-year rule of thumb generally is accepted as the minimal life-expectancy prerequisite for considering curative local treatment in men with PCa. Unfortunately, life tables are not reliable for predicting life expectancy in definitive therapy candidates and demonstrate only 60% accuracy.86 Moreover, clinicians also are poor raters of life expectancy, and 19 clinicians yielded only 69% accuracy. To circumvent this problem, Tewari et al (n = 3159 patients),87 Cowen et al (n = 506 patients),88 and Albertsen et al (n = 451 patients)43, 89 developed models that predict life expectancy in men with PCa. The accuracy of these models ranged from 69% to 73%. Unfortunately, none of the 3 models was capable of discerning between PCa-specific and all-cause mortality. To address these considerations, Walz et al devised a nomogram for predicting life expectancy in excess of 10 years in RP and external-beam radiotherapy candidates (n = 9131 patients).90 Their model focused on patients without evidence of recurrent cancer after definitive therapy. Its predictors consist of only age and comorbidities. Internal validation of this nomogram yielded 84% accuracy versus 81% for the model described by Tewari et al. On the basis of accuracy, simplicity, and generalizability, the model of Walz et al appears to provide the best estimates of life expectancy in excess of 10 years.

Table 7. Prediction of Life Expectancy in Men With Prostate Cancer

Reference	Prediction form	Patient Population	Outcome	No. of Patients	Variables	Accuracy, %	Validation
RP indicates radical prostatectomy; RT, radiotherapy,
Albertson 199689	Probability formula	Clinically localized prostate cancer	Overall survival, 10 y	451	Age, Gleason sum, index of coexistent disease category	71	Not performed
Tewari 200487	Probability graph	Clinically localized prostate cancer	Overall survival, 10 y	6149	Age, race, comorbidity, PSA, Gleason sum, treatment type	63	Not performed
Cowen 200688	Probability nomogram development	Clinically localized prostate cancer	Life expectancy, 5-15 y	506	Age, Charlson comorbidity index, presence of angina, systolic blood pressure, body mass index, smoking, marital status, PSA, Gleason sum, clinical stage, treatment type (RP vs RT vs other)	73	Internal
Walz 200790	Probability nomogram development	Clinically localized prostate cancer	Life expectancy, 10 y	9131 (RP, 5955; external-beam RT, 3176)	Age, Charlson comorbidity index, treatment type (RP vs external-beam RT)	84.3	Internal

Nomograms of the Future—Inclusion of Novel Biomarkers and Imaging Tools

The accuracy of current predictive tools still is not perfect. To date, the addition of other potentially informative clinical and pathologic features has not improved the accuracy of these models significantly.91, 92 The incorporation of novel biomarkers and/or imaging tools that are associated with the biologic behavior of PCa potentially may improve predictive accuracy (Table 8). Despite numerous reports of promising new biomarkers in the urologic literature, to our knowledge only 3 studies to date have demonstrated a statistically significant improvement in predictive accuracy when biomarkers were added to established predictors in the nomogram setting (for example, see Fig. 6).33, 93, 94

Figure 6. Pretreatment nomogram for predicting 5-year biochemical disease recurrence-free probability after radical prostatectomy, including preoperative plasma levels of transforming growth factor β1 (TGFβ1) and interleukin-6 soluble receptor (IL6R). PSA indicates prostate-specific antigen; Bx, biopsy; GG, Gleason grade; Preop, preoperative; Prog Free Prob, progression-free probability. Reproduced with permission from Kattan MW, Shariat SF, Andrews B, et al. The addition of interleukin-6 soluble receptor and transforming growth factor beta1 improves a preoperative nomogram for predicting biochemical progression in patients with clinically localized prostate cancer. J Clin Oncol. 2003;21:3573-3579. Reprinted with permission. ©2008 American Society of Clinical Oncology. All rights reserved.

Table 8. Prediction of Specific Pathologic Features or Biochemical Recurrence in Men Who Underwent Radical Prostatectomy for Clinically Localized Prostate Cancer Based on Novel Variables

Reference	Prediction Form	Operative Status	Novel Variable(s)	Outcome	No. of Patients	Variables	Accuracy, %	Validation
MRI indicates magnetic resonance imaging; MRSI, magnetic resonance spectroscopic imaging; LN, lymph node; PSA, prostate-specific antigen; TGF-β1, transforming growth factor β1; IL-6, interleukin 6.
Wang 200650	Probability nomogram development	Preoperative	MRI and MRSI	LN invasion	411	MRI variables and Partin probability table (Partin 199719)	89	Internal
Wang 2007^549	Probability nomogram development	Preoperative	MRI and MRSI	Seminal vesicle invasion	573	MRI variables and preoperative Kattan probability nomogram (Kattan 199810)	87	Internal
Shukla-Dave 200795	Probability nomogram development	Preoperative	MRI and MRSI	Insignificant cancer (organ-confined cancer ≤0.5 cm^3 with no poorly differentiated elements)	220	Pretreatment PSA, clinical stage, biopsy cores positive, pretreatment MRI volume of prostate, and overall MRI/MRSI score	85	Internal
Kattan 200333	Probability nomogram development	Preoperative	Plasma levels of TGF-β1 and IL-6 soluble receptor	Biochemical recurrence	714	Preoperative plasma TGF-β1 and IL-6 soluble receptor and preoperative Kattan probability nomogram (Kattan 199810)	83	Internal
Stephenson 200593	Probability nomogram development	Postoperative	Gene expression based on oligonucleotide microarrays	Biochemical recurrence	79	Gene expression signatures and postoperative Kattan probability nomogram (Kattan 199911)	89	Internal
Shariat 200794	Probability nomogram development	Preoperative	Plasma levels of plasminogen activator inhibitor 1	Biochemical recurrence	429	Preoperative plasma plasminogen activator inhibitor 1 and preoperative Kattan probability nomogram (Kattan 199810)	79	Internal

Noninvasive diagnostic imaging, especially magnetic resonance imaging and magnetic resonance spectroscopic imaging, has improved in recent years and is gaining widespread acceptance for aiding PCa diagnosis, tumor localization, staging, assessment of tumor aggressiveness, and treatment planning. Investigators have used nomogram and neural network modeling to predict organ-confined PCa,49 clinically significant disease,95 and biochemical recurrence after RP.96

DISCUSSION

The paucity of randomized trials in PCa makes diagnostic and treatment decisions complex. Decision rules such as nomograms provide evidence-based and, at the same time, individualized predictions of the outcome of interest. It has been demonstrated repeatedly that such predictions are more accurate than those of clinicians, regardless of their level of expertise. Nomograms have been embraced by the urologic community. The nomogram format also has been adopted in several other disciplines of oncology, such as breast, colon, bladder, gastric, and lung cancers.

The exponential growth of new nomograms prompted some criticisms. The fundamental issue raised by critiques of predictive tools is their usefulness. Indeed, limited data exist with respect to the impact of predictive tools on medical decision making. It has been demonstrated that decision aids, which can include prediction models, improve patient knowledge and affect decision-making behavior97; however, to our knowledge the actual patient outcome after the intervention of a predictive tool is unknown. In this respect, the role of predictive tools has yet to be proven. A clinical trial that established the effect of nomograms on patient medical decision making would be very valuable; however, informing patients with predictions regarding the impact of a medical procedure appears to be ethical, whereas withholding accurate outcome predictions from patients to achieve equipoise in a randomized trial in which the control arm lacks this information does not appear to be ethical. Currently, patients are using very limited information when making their decisions, and direct outcome predictions are the simplest factors for them to consider; decision making is facilitated when patients can see tailored predictions of their outcomes with various alternatives.

In addition to improved medical decision making, accurate risk estimates are also required for the evaluation of novel markers33, 98 and clinical trial design to ensure that homogeneous, high-risk patient groups are used to investigate new cancer therapeutics. Prediction models have the potential of improving the ability of phase 2 trials to discriminate between ineffective and potentially effective therapy. As in the majority of phase 2 trials, both the expected response rate from standard care and the actual response rate of novel therapy are group-level estimates: The researchers simply take the number of patients who respond and divide by the total treated to get a proportion. This takes no account of individual patient characteristics and, thus, implicitly assumes similarity of trial patients and historic controls. However, trial patients may differ from controls in terms of performance status, prior experience of chemotherapy, extent of disease, or other prognostic factors. Apparently promising or disappointing results from a phase 2 treatment trial merely may be the result of such differences.99 For phase 3 trials, prediction models can help to ensure that eligible patients are at sufficient levels of high risk, thereby increasing event rates and reducing sample size requirements. Risk-prediction models define high-risk patients more accurately than risk-grouping strategies. The use of risk predictions for individual patients, therefore, decreases the proportion of low-risk patients enrolled, avoiding unethical inclusion and increasing statistical power. Finally, future designs of phase 3 trials should include prediction models to increase the clinical utility of their findings.

The results of such trials typically are applied by citing the average group-level effect. However, the use of a risk-prediction model may indicate that a patient's true risk was substantially different from the group-level estimate. This could lead a patient to make a different decision regarding treatment. Predictions models, as discussed above, have greater value to the clinician for counseling individual patients, because prediction models discriminate between high-risk and low-risk patients more accurately than traditional subgroups, which are based only on risk factors. Therefore, we recommend the wider adoption of risk-prediction models in the design, analysis, and implementation of clinical trials. The electronic versions of many of the nomograms described in this report are available online at www.nomograms.org or at www.nomogram.org, accessed on January 12, 2008. On both sites, the tools are organized according to the disease stages of PCa.

Many more nomograms, as well as improvements to existing nomograms, are needed. For example, none of the nomograms predicts with perfect accuracy. Novel biomarkers, larger datasets, systematic and clean data collection, and more sophisticated modeling procedures are needed to improve predictive accuracy. Provider-specific or center-specific models may possibly increase the specificity and accuracy of nomograms further.

Conclusions

Despite their limitations and the limitations of data, predictive tools are difficult to beat for individualized, evidence-based medical decision making. Increasingly, patients, administrators, peers, and third-party payers demand to have an objective justification for nearly all clinical decisions. Evidence-based predictive tools can provide individualized estimates for several endpoints in urologic oncology. Moreover, when determining the usefulness of nomograms, many patients want to know their likely outcomes, and most clinicians would prefer to provide accurate estimates of those outcomes. When a nomogram is available, little else is able to make more accurate predictions of outcome, and this confirms the usefulness of nomograms. Predictive tools have empowered patients and physicians in their fight against PCa by providing highly accurate, reproducible, evidence-based, yet individualized disease-related risk estimations that facilitate management-related decisions.